ikuznet1 / data-analytics-website Goto Github PK

find some articles to read about it, i'll help you filter them.
laura balzano has written some stuff
rob nowak
becca willet
ben recht

are the most readable ones i know.

eg

• nan's/infs break code
• rows should be data points, columns should be features (or vice versa)
• appropriate size for things to work

• 1
• asdf
• [x]

• Color scheme setting each page?

• Something like scree for tsne
• Better caching for tsne
• Find std of R-scores
• Allow for boxplot scaling
• Consider using scalefillbrewer(type = 'div') for heatmaps
• Scree plots look wrong
• Add kmeans++ clustering
• Check if outliers removed from embed

not sure how.
play with how matlab does it for importing.
we can use default R impute options as one option?

• cluster on embedded points.
• choose # of dimensions via Zhu & Godsi: https://uwaterloo.ca/data-science/sites/ca.data-science/files/uploads/files/automatic-dimensionality-selection-from-scree-plot-use-of-profile-likelihood.pdf (getElbow.R lives somewhere, ask greg or youngser for it)
• we can later enable user to choose the dimension of embedding.

• on "Data" tab, show: # features, # samples, nnz, # NaN
• on "Marginal" tab, make "combined" the default.
• throughout, rename tabs to be more informative
• "Distance Metric" = "Euclidean Distance"?
• fix size of correlation matrix
• mean vector plotting doesn't really have scatter, probably we want 'jittered scatter'
• add iris.csv file for demo purposes, you can leave the other one, maybe name it something more informative.